Yibing Liu

Bispectrum analysis for feature extraction of pitting fault in wind turbine gearbox

This paper discusses the theory of higher-order statistical analysis and its application in gear pitting fault feature extraction from gearbox vibration signals analysis of a large scale wind turbine generator system (WTGS). The bispectrum was used to inhibit the Gaussian noise in measured vibration signals and to reveal the fault related non-Gaussian information. We propose to divide the dual-frequency plan of bispectrum into several partitions and use the total amplitude value of each partition, which related to the non-Gaussian intensity of vibration signals, as feature values for identification of pitting fault. It can be seen by comparing the results between pitting fault and normal condition that the proposed method are effective for the extraction of gear pitting fault information from noised vibration signals and bring stable performance, high sensitivity.

Kernel Principal Components Analysis for Early Identification of Gear Tooth crack

In the application of the fault diagnosis, principal components analysis (PCA) is often used to judge the state of a equipment and classify the faults by means of projecting the original data to the principal components space. However, if the data are concentrated in a nonlinear subspace, PCA will fail to work well. Kernel principal components analysis (KPCA) transforms the input data from the original input space into a higher dimensional feature space with the nonlinear mapping, and then uses the nonlinear principal components to classify the state of the equipment. In this paper a case of gear fault diagnosis was studied by KPCA. The feature value was firstly extracted from vibration signals of the gearbox under the condition of continue running, and then KPCA method was used to extract the information of gear crack fault. The result shows that KPCA can be more effective to distinguish the state of the gear and more suitable to diagnose the gear faults in early stage

Early Identification of Machine Fault Based on Kernel Principal Components Analysis

Principal Components Analysis (PCA) is used to classify the running condition of a machine by means of projecting the original data to the Principal Components space. However, if the data are concentrated in a nonlinear subspace, PCA will fail to work well. Kernel Principal Components Analysis (KPCA) transforms the input data from the original input space into a higher dimensional feature space with the nonlinear mapping, and then uses the nonlinear principal components to realize the classification. In this paper a case of gear fault diagnosis was studied with KPCA. The characteristic values of frequent domain from vibration signals of the gearbox under the running condition were extracted, and the KPCA method was used to classify gear crack fault. The result shows that KPCA is more effective to distinguish the state of the gear and more suitable to diagnose the gear faults in early stage.

Amplitude envelope analysis for feature extraction of direct-driven wind turbine bearing failure

Made use of the vibration signals of main shaft which were measured from a normal direct-driven wind turbine and a fault one, the vibration characteristics of main shaft bearing were analyzed and studied. From time domain, frequency domain and cepstrum domain, analyzed and compared the vibration signal characteristics between normal and fault units, then narrowband envelope analysis had been made. The difference of envelope spectrum for vibration signals between normal and fault units was revealed. The results show that narrowband envelope analysis can effectively identify the bearing fault state.

Turbine vibration source separation based on Independent Component Analysis

Independent Component Analysis (ICA) is a new method for solving the blind source separation (BBS) problem, and has been widely used in recent years in many industrial areas to decompose mixed signals into several mutual independent components. In practical work, many source mixing models are not linear model but convolution model, for which the basis linear ICA model is not suitable. In this paper we propose to use frequency domain ICA (FDICA) model to solve those problems with convolution mixing model. The observed multi-channel signals are firstly transformed to frequent domain by means of Fourier Transform, so that the convolution mixing model is changed to a linear mixing model, which can be separated with linear ICA model. A case study is showed with measured signal from the shaft vibration of a large steam turbo-generator set in a power plant. The separation with FDICA for varies signal types are discussed. The result shows that frequency domain method of ICA can separate the mixed signal clearly. The result of separation is stable in regular condition.

Application of over-complete ICA in separating turbine vibration sources

Over-complete ICA problem are always met in engineering applications. That is to say, the number of unknown sources is more than the number of observed signals. At this time basic ICA model is not suitable. This text utilizes the component of priori knowledge as additional input signal (addition virtual channel), to increase the number of the input signals. And it can solve the engineering application problem of over-complete ICA. This method is tested through a group of actual turbine vibration signals. The similarity coefficient is introduced to verify the effect of source separation.

Analysis of lower frequent vibration of large wind turbine tower

Tower of modern large wind turbine are made of flexible steel tubular structure. Its first nature frequency is normally lower than the blade pass frequency and even lower than the wind rotor rotating frequency, therefore the vibration feature of the tower is important for the running condition of the wind turbine set. This paper discusses the lower vibration feature of the tower of a large MW wind turbine by means of analyzing the measured tower vibration signal. The non-stationary vibration feature was discussed based on modern spectral analysis and time-frequent analysis. The lower frequent resonance of the tower was identified

Feature extraction of turbine abnormal vibration condition based on ICA

Independent component analysis of a single measured mixing signal, that is single channel Independent component analysis (SCICA), has been widely used in feature extraction of a signal. In this paper we provide a example of using single channel ICA for extracting the feature of a abnormal running condition of a turbine from measured vibration signals, in order to show the effect of SCICA. The bearing vibration signals of the turbine are measured under two different running conditions. We choose one bearing vibration signal with clear abnormal information to test. The results show that the basic functions of SCICA have different characters for different turbine running condition and can be used as features. The results need further analyze to distinguish the different states.

Feature extraction of gear tooth crack based on the method of surrogate data

Nonlinear characteristics always exist in mechanical systems, in particular, if faults occur in mechanical systems. Therefore the nonlinear component may be used to identify the faults. The primary goal of this paper was to exact the feature of gear tooth crack by means of the method of surrogate data from nonlinear dynamics using real vibration signal measured on a gear test device. Surrogate data were generated from a simpler linear model and had similar statistical characteristics with the original data. The mutual information was applied as the statistic to determine if the original data were significantly different from its surrogates. The vibration signal was divided into several sections. Each sections failure may be identified through computing surrogates significant levels. The results showed that the different gear running states had different significant levels of surrogate data, and the changes of running states had the same trend with the increase process of gear crack. It was concluded that the significant level could be used as the numerical basis for fault diagnosis and gear case analysis, which was conducive to the further analyze the fault development.